Supplier credit assessment method based on multi-source heterogeneous data and related device

Abstract

The embodiment of the invention relates to the field of data processing, and provides a supplier credit assessment method based on multi-source heterogeneous data and a related device, and the method comprises the steps: obtaining a historical multi-source heterogeneous data set related to a to-be-assessed supplier; based on a preset supplier credit assessment exclusive knowledge base, performing dynamic logic calibration processing on each piece of historical multi-source heterogeneous data in the historical multi-source heterogeneous data set to obtain a calibrated multi-source heterogeneous data set; according to the calibrated multi-source heterogeneous data set, carrying out credibility-based multi-modal fusion processing to obtain a supplier double-weight fusion feature data set; performing dynamic state updating processing according to the supplier double-weight fusion feature data set to obtain a supplier dynamic feature data set; and according to the supplier dynamic feature data set, supplier credit assessment processing is performed on the to-be-assessed supplier to obtain a target supplier credit assessment result, so that the accuracy of a credit assessment process for the supplier can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a supplier credit assessment method and related apparatus based on multi-source heterogeneous data. Background Technology

[0002] In supply chain management and business cooperation decision-making, supplier credit assessment is a crucial step in managing cooperation risks and ensuring supply chain stability. With the increasing complexity of the market environment and the diversification of supply chain networks, enterprises are demanding higher accuracy in supplier credit assessments. However, existing supplier credit assessment technologies have limited capabilities in processing multi-source heterogeneous data, facing challenges such as difficulty in overcoming data format barriers, unresolved semantic ambiguities, and untimely conflict correction. These issues directly affect the quality of the assessment basis, resulting in inaccurate final credit assessment results. Therefore, improving the accuracy of the supplier credit assessment process has become an urgent problem to be solved. Summary of the Invention

[0003] This application provides a supplier credit assessment method and related apparatus based on multi-source heterogeneous data, which can effectively improve the accuracy of the supplier credit assessment process.

[0004] The first aspect of this application provides a supplier credit assessment method based on multi-source heterogeneous data, which includes: Obtain historical, multi-source, heterogeneous datasets related to the suppliers to be evaluated; Based on a pre-defined supplier credit assessment knowledge base, dynamic logical calibration is performed on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset. Based on the calibration of the multi-source heterogeneous dataset, a confidence-based multimodal fusion process is performed to obtain the supplier dual-weighted fusion feature dataset; The supplier dynamic feature dataset is obtained by performing dynamic state update processing on the supplier dual-weight fusion feature dataset; Based on the supplier dynamic feature dataset, the supplier credit assessment is performed on the supplier to be evaluated to obtain the target supplier credit assessment result.

[0005] A second aspect of this application provides a supplier credit assessment device based on multi-source heterogeneous data, the supplier credit assessment device based on multi-source heterogeneous data comprising: The acquisition unit is used to acquire historical multi-source heterogeneous datasets related to the suppliers to be evaluated. The first processing unit is used to perform dynamic logical calibration processing on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset based on a preset supplier credit assessment exclusive knowledge base, so as to obtain a calibrated multi-source heterogeneous dataset. The second processing unit is used to perform confidence-based multimodal fusion processing on the calibration multi-source heterogeneous dataset to obtain a supplier dual-weight fusion feature dataset. The third processing unit is used to perform dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain the supplier dynamic feature dataset. The fourth processing unit is used to perform supplier credit assessment processing on the supplier to be evaluated based on the supplier dynamic feature dataset, and obtain the target supplier credit assessment result.

[0006] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the first aspect of this application.

[0007] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of this application.

[0008] A fifth aspect of this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of this application. The computer program product may be a software installation package.

[0009] Implementing the embodiments of this application has the following beneficial effects: By acquiring historical multi-source heterogeneous datasets related to the suppliers to be evaluated, and based on a pre-defined supplier credit assessment knowledge base, dynamic logical calibration processing is performed on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset. Further, based on the calibrated multi-source heterogeneous dataset, a credibility-based multimodal fusion processing is performed to obtain a supplier dual-weight fusion feature dataset. This allows for dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain a supplier dynamic feature dataset. Finally, based on the supplier dynamic feature dataset, supplier credit assessment processing is performed on the suppliers to be evaluated, resulting in more accurate target supplier credit assessment results, which helps improve the accuracy of the supplier credit assessment process. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This application provides a schematic diagram of the structure of a supplier credit assessment method based on multi-source heterogeneous data. Figure 2 This application provides a flowchart illustrating a supplier credit assessment method based on multi-source heterogeneous data. Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of the structure of a supplier credit assessment device based on multi-source heterogeneous data. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0014] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0015] To better understand the supplier credit assessment method based on multi-source heterogeneous data provided in this application, a brief introduction to existing supplier credit assessment methods is given below. First, existing technologies have limited processing capabilities for multi-source heterogeneous data. Supplier credit-related data often originates from multiple channels, including internal enterprise systems, third-party credit reporting platforms, and industry association announcements. Data formats encompass structured financial statements, semi-structured contract texts, and unstructured qualification certificate images. However, traditional assessment methods often use a single-format data input or simply integrate heterogeneous data through simple format conversion, failing to effectively calibrate for the differences in data sources. This results in unresolved issues such as ambiguous terminology and logical conflicts within the data, directly affecting the reliability of the assessment basis. Second, existing technologies generally lack mechanisms for quantifying and integrating data credibility. Data from different sources varies significantly in authority, timeliness, and relevance; for example, the credibility of financial data issued by financial institutions is incomparable to that of anonymous online evaluations. However, traditional methods often treat all types of data equally, failing to assign appropriate weights based on data credibility. This leads to low-quality data interfering with the assessment results, making it difficult to accurately reflect the supplier's true creditworthiness. Furthermore, existing assessment processes lack dynamic updating capabilities. Since a supplier's operational status, performance capabilities, and other credit-related characteristics change dynamically over time, traditional assessment methods often rely on static historical data for one-time evaluations. This fails to incorporate the supplier's latest dynamic information (such as recent delivery records and financial fluctuations), resulting in assessment results that are outdated and ill-suited to the dynamic market environment and cooperation needs. Moreover, the integration logic of existing assessment dimensions is not scientifically sound. Some methods focus only on a single dimension (e.g., solely based on performance records or financial data), ignoring the synergistic effect of multiple dimensions. Even when multi-dimensional assessments are used, they often employ fixed weight allocation methods without dynamically adjusting based on the actual importance of features and data quality, leading to insufficient comprehensiveness and relevance in the assessment results.

[0016] To address the aforementioned issues, this application provides a supplier credit assessment method based on multi-source heterogeneous data. This method leverages a pre-defined supplier credit assessment knowledge base to standardize and dynamically correct conflicts within the heterogeneous data. It overcomes data format barriers and ensures feature quality through a credibility-based multimodal dual-weight fusion mechanism. Furthermore, it incorporates the latest supplier status information in a real-time dynamic update process, optimizing the entire process from data processing to assessment output. Ultimately, this achieves a comprehensive, accurate, and dynamic assessment of supplier credit, effectively improving the accuracy of supplier credit assessment and providing scientific and reliable technical support for enterprise cooperation decisions and supply chain risk management.

[0017] Please see Figure 1 , Figure 1A schematic diagram of a supplier credit assessment system based on multi-source heterogeneous data is shown. Figure 1 As shown, a supplier credit assessment system based on multi-source heterogeneous data can include a data acquisition module, a dynamic logical calibration module, a multi-modal fusion module, a dynamic status update module, and a supplier credit assessment module. The data acquisition module can collect structured, semi-structured, and unstructured multi-source heterogeneous data from internal and external channels. The dynamic logical calibration module can extract key information, map terms, verify correlations, and perform hierarchical correction to output a calibrated multi-source heterogeneous dataset. The multi-modal fusion module can quantify the credibility of the calibration data, extract and integrate features from different modalities, and perform semantic alignment, classification, weighted fusion, and weight allocation to output a supplier dual-weighted fused feature dataset. The dynamic status update module can separate static features from historical dynamic features and extract real-time dynamic features from real-time acquired data, performing timeliness detection and freshness fusion to output a supplier dynamic feature dataset. The supplier credit assessment module can generate credit scores, levels, and risk warnings based on the supplier dynamic feature dataset using a preset model, providing a basis for decision-making.

[0018] Please see Figure 2 , Figure 2 This application provides a flowchart illustrating a supplier credit assessment method based on multi-source heterogeneous data. For example... Figure 2 As shown, the supplier credit assessment method based on multi-source heterogeneous data includes: S10: Obtain historical multi-source heterogeneous datasets related to the suppliers to be evaluated.

[0019] Among them, the supplier to be evaluated can refer to the enterprise or organization that needs to undergo credit qualification review through this method, that is, the target object that needs to be credit evaluated based on its multi-source heterogeneous data.

[0020] Historical multi-source heterogeneous datasets may include one or more historical multi-source heterogeneous datasets. These datasets can refer to data collected from different sources (such as corporate websites, third-party credit reporting platforms, supply chain management systems, etc.) within a past period, and can be in various formats. These historical multi-source heterogeneous datasets may include structured data (such as financial statements), semi-structured data (such as contract texts), and unstructured data (such as images of qualification certificates), etc., and this application does not impose any restrictions on this.

[0021] Specifically, the operating data, performance records, compliance information, etc. of the suppliers to be evaluated in recent years (such as the past 1-3 years) can be collected through methods such as application programming interface (API) docking, web scraping, or manual input to form the original dataset, which is the aforementioned historical multi-source heterogeneous dataset. This application does not impose any restrictions on this.

[0022] S20: Based on the preset supplier credit assessment knowledge base, perform dynamic logical calibration processing on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset.

[0023] Among them, the pre-set supplier credit assessment knowledge base can refer to a professional database that includes a terminology database specific to supplier credit assessment, logical verification rules (such as "settled orders should not have outstanding payment records"), and dynamically updated compliance standards (such as the latest supply chain management regulations).

[0024] Dynamic logical calibration can be understood as a process of standardizing terminology and correcting logical conflicts in the original data (i.e., historical multi-source heterogeneous data). Understandably, while dynamic logical calibration may retain the heterogeneity of data formats (such as tables, text, and images), it can eliminate ambiguity in the content of historical multi-source heterogeneous data.

[0025] The calibration multi-source heterogeneous dataset may include one or more calibration multi-source heterogeneous data sets, which may refer to data after dynamic logical calibration. That is to say, the format of the calibration multi-source heterogeneous data in the calibration multi-source heterogeneous dataset can still be diverse, but the key information is uniformly expressed and logically consistent, and this application does not impose any restrictions on this.

[0026] Specifically, key information, such as delivery delays and debt-to-equity ratios, can be extracted from the raw data to map heterogeneous terms to standard terms based on a knowledge base. For example, "supply delay → delivery postponement." Knowledge base rules can then be used to verify the correlation between data, such as whether order amounts match invoice amounts, thereby correcting conflicting data to obtain the calibrated multi-source heterogeneous dataset. It should be noted that the details of dynamic logic calibration processing are described in the following section and will not be repeated here.

[0027] S30: Perform confidence-based multimodal fusion processing on the calibration multi-source heterogeneous dataset to obtain a supplier dual-weight fusion feature dataset.

[0028] Among them, credibility-based multimodal fusion processing can be understood as the process of combining the credibility of data sources (such as the credibility of financial institution data being higher than that of anonymous evaluations) to transform data of different formats into unified features and assign weights.

[0029] The supplier dual-weighted fusion feature dataset may include one or more supplier dual-weighted fusion feature data, which may refer to the feature data after multimodal fusion. Optionally, each supplier dual-weighted fusion feature data may carry individual credibility weight (i.e., data quality itself) and dimension category weight (i.e., importance in the credit assessment process), which is not limited in this application.

[0030] Specifically, the credibility of various data sources can be quantified. Specific quantification indicators can include authority, freshness, and correlation with credit dimensions. Furthermore, features can be extracted modally, such as numerical features from structured data, semantic features from text, and key information from images. This allows for the weighted merging of similar features based on credibility; for example, a weighted average can be taken for on-time delivery rates from different sources. Dissimilar features retain their original credibility. Then, based on the importance of each dimension category in the credit assessment process, corresponding weights are assigned to each feature to achieve dual-weight labeling for each feature. It should be noted that the relevant content of credibility-based multimodal fusion processing can be found in the detailed description below, and will not be repeated here.

[0031] S40: Perform dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain the supplier dynamic feature dataset.

[0032] The dynamic status update process can be understood as supplementing with real-time data and replacing historical features to ensure that the final features reflect the supplier's latest status. The supplier dynamic feature dataset can include one or more supplier dynamic feature data sets, which can refer to feature data that dynamically reflects the supplier's most accurate and up-to-date status. Optionally, the supplier dynamic feature dataset can include static basic feature data (such as registered capital) and dynamic feature data that has been updated in real-time (such as delivery status in the last 3 days).

[0033] Specifically, the dual-weighted fusion feature data of each supplier can be split into static slices (long-term unchanging features) and historical dynamic slices (features that have changed in the last 3 months). New data is collected in real time, and after calibration and fusion, real-time dynamic slices are extracted. Time-sensitive historical dynamic slices are then further filtered out (e.g., using a sliding window). These time-sensitive historical dynamic slices are then fused with the real-time dynamic slices according to their freshness (real-time data has a higher weight). This process is further integrated with the static slices and the updated fused dynamic slices to obtain the aforementioned supplier dynamic feature dataset. This application does not impose any limitations on this process. It should be noted that the relevant content regarding dynamic state update processing can be found in the detailed description below, and will not be repeated here.

[0034] S50: Perform supplier credit assessment processing on the supplier to be evaluated based on the supplier dynamic feature dataset to obtain the target supplier credit assessment result.

[0035] The credit assessment process can be understood as the process of further determining the supplier's credit rating based on a dynamic feature dataset of the supplier. Optionally, a model, such as a machine learning model or a rule engine, can be used to determine the supplier's credit rating based on the dynamic feature dataset of the supplier; this application does not impose any restrictions on this. The target supplier credit assessment result can refer to a credit assessment result that includes information such as the supplier's credit assessment score, supplier credit assessment level (e.g., AAA), and risk warnings (e.g., medium performance risk).

[0036] Specifically, the supplier dynamic feature dataset can be input into a preset credit assessment model to invoke the weight configuration rules for each assessment dimension (such as performance capability, financial health, compliance reputation, etc.). For example, the credibility weight and dimension category weight carried by each supplier's dynamic feature data can be combined to perform weighted calculations on the sub-features under each dimension. For instance, in the performance capability dimension, the scores of sub-features such as the latest on-time delivery rate (credibility weight 0.9 × dimension category weight 40%) and production equipment utilization rate (credibility weight 0.85 × dimension category weight 30%) can be proportionally summarized to obtain the comprehensive score for each dimension. Subsequently, based on the pre-defined model... The system assigns weights to specific dimensions (e.g., 40% for performance capability, 35% for financial health, and 25% for compliance reputation) and weights the scores of each dimension to obtain the supplier's total credit score. Finally, it uses a pre-defined score-level mapping rule (e.g., AAA for scores above 90, AA for scores between 80 and 89) to convert the total credit score into a corresponding credit level. A risk analysis report is then generated based on the differences in scores across each dimension. For example, a low score in the financial health dimension could raise concerns about the risk of a broken cash flow; excellent performance in performance capability could lead to priority consideration for long-term cooperation. This allows for precise reference in subsequent steps based on the target supplier's credit assessment results, providing a basis for cooperation decisions.

[0037] By selectively collecting historical, multi-source, heterogeneous datasets of suppliers to be evaluated, eliminating data ambiguity through dynamic logical calibration, overcoming heterogeneous data format barriers and ensuring feature quality through credibility-based multimodal fusion processing, and ensuring the timeliness of the evaluation basis through dynamic status update processing, a comprehensive and accurate evaluation of supplier credit is ultimately achieved. This effectively solves the problems of messy data, low credibility, and delayed results in traditional evaluations, providing scientific and reliable support for the collaborative decision-making and risk management of the evaluation subjects.

[0038] In this embodiment, by acquiring historical multi-source heterogeneous datasets related to the supplier to be evaluated, and based on a preset supplier credit assessment-specific knowledge base, dynamic logical calibration processing is performed on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset. Further, based on the calibrated multi-source heterogeneous dataset, a credibility-based multimodal fusion processing is performed to obtain a supplier dual-weight fusion feature dataset. This allows for dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain a supplier dynamic feature dataset. Finally, based on the supplier dynamic feature dataset, supplier credit assessment processing is performed on the supplier to be evaluated, resulting in a more accurate target supplier credit assessment result, which helps improve the accuracy of the supplier credit assessment process.

[0039] In one possible implementation, dynamic logical calibration can be performed through four steps: extracting key information, unifying terminology mapping, verifying relevance, and hierarchical correction. For example, first, key information is extracted from historical multi-source heterogeneous data; then, terminology is unified based on a dedicated knowledge base; subsequently, the logical relevance between information is verified; and finally, conflicting data is hierarchically corrected, thereby obtaining a calibrated multi-source heterogeneous dataset. Specifically, a method for performing dynamic logical calibration on each historical multi-source heterogeneous data point in the historical multi-source heterogeneous dataset based on a pre-defined supplier credit assessment dedicated knowledge base to obtain a calibrated multi-source heterogeneous dataset may include: A1. Extract key information from each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a key information dataset. A2. Based on a pre-set supplier credit assessment knowledge base, perform term mapping processing on each key information data in the key information dataset to obtain a standard key information dataset. A3. Call the logical rules in the preset supplier credit assessment knowledge base to perform correlation verification processing on each standard key information data in the standard key information dataset to obtain a correlation conflict dataset. A4. Perform hierarchical dynamic correction processing on each related conflict data in the related conflict dataset to obtain a calibrated multi-source heterogeneous dataset.

[0040] The key information dataset may include one or more key information data, which may refer to core information related to credit assessment extracted from historical multi-source heterogeneous data, such as a 3-day delivery delay or a debt-to-asset ratio of 60%, etc. This application does not impose any restrictions on this.

[0041] Specifically, technologies such as optical character recognition (for image data), natural language processing (for text data), and field extraction (for structured data) can be used to filter out information related to credit dimensions such as performance capability and financial status from various historical data, thereby further summarizing and forming the aforementioned key information dataset.

[0042] The standard key information dataset can include one or more standard key information items. These standard key information items refer to key information that uses a unified expression after terminology mapping. Specifically, each key information item in the key information dataset can be compared and matched with standard terms in a pre-defined supplier credit assessment knowledge base. This standardizes and transforms synonymous but heterogeneous terms (such as payment cycle and payment period) to ensure that all key information is expressed consistently, thereby forming and obtaining the aforementioned standard key information dataset.

[0043] Logical rules can refer to the criteria and norms stored in a pre-defined supplier credit assessment knowledge base, used to verify the logical rationality between various standard key information, such as rules that settled orders should not have outstanding payment records, and that the number of delivery delays is positively correlated with the number of contract defaults. The correlation conflict dataset includes one or more correlation conflict data points, which can refer to information found to be logically contradictory through verification. For example, the simultaneous existence of "order status is completed" and "outstanding payment amount is 500,000 yuan," etc., is not limited in this application.

[0044] Specifically, logical rules in the knowledge base can be invoked to cross-validate information in the standard key information dataset, such as verifying the correlation between financial data and performance records, in order to filter out conflicting information that violates the logical rules, thereby forming and obtaining the aforementioned correlation conflict dataset.

[0045] The calibration multi-source heterogeneous dataset can include one or more calibration multi-source heterogeneous datasets. These calibration multi-source heterogeneous datasets can refer to multi-source heterogeneous datasets that have undergone hierarchical dynamic correction, have consistent terminology, are logically consistent, and meet the evaluation requirements.

[0046] Specifically, conflicts can be categorized according to their severity (e.g., minor conflicts: expression errors; severe conflicts: data contradictions). By combining the real-time updated rules (e.g., the latest industry standards) and historical correction records in the pre-set supplier credit assessment knowledge base, corresponding correction measures can be taken for conflict data of different levels. For example, minor conflicts can be automatically corrected according to standard rules, while severe conflicts can trigger manual review and synchronous updates to the knowledge base, in order to further form and obtain the above-mentioned calibrated multi-source heterogeneous dataset.

[0047] In this embodiment, by extracting key information, mapping standardized terms, verifying accurate correlations, and performing hierarchical dynamic correction, semantic ambiguities and logical conflicts in historical multi-source heterogeneous data are effectively eliminated. Relying on the real-time update characteristics of the pre-set supplier credit assessment knowledge base, the dynamic adaptability of data processing is ensured. The final output calibrated multi-source heterogeneous dataset has the characteristics of unified terminology, rigorous logic, and reliable quality, providing solid data support for subsequent credibility-based multimodal fusion processing and accurate supplier credit assessment, and significantly improving the scientificity and accuracy of the entire credit assessment process.

[0048] In one possible implementation, during the credibility-based multimodal fusion processing, the credibility of each data point in the calibration multi-source heterogeneous dataset is first quantified to obtain a credibility score set. Then, structured numerical features, semi-structured semantic features, and unstructured key features are extracted from the calibration multi-source heterogeneous dataset to form three feature data subsets. These three subsets are then integrated to obtain the supplier multimodal feature dataset. Finally, the dataset is weighted and fused using the credibility score set to obtain the supplier dual-weighted fusion feature dataset. Specifically, a method for performing credibility-based multimodal fusion processing on the aforementioned calibration multi-source heterogeneous dataset to obtain the supplier dual-weighted fusion feature dataset may include: B1. Perform confidence quantification on each calibration multi-source heterogeneous data in the calibration multi-source heterogeneous dataset to obtain a set of confidence scores; B2. Perform structured numerical feature extraction processing on the aforementioned calibrated multi-source heterogeneous dataset to obtain a subset of structured numerical feature data; B3. Perform semi-structured semantic feature extraction processing on the aforementioned calibrated multi-source heterogeneous dataset to obtain a subset of semi-structured semantic feature data; B4. Based on the calibration multi-source heterogeneous dataset, perform unstructured key feature extraction processing to obtain a subset of unstructured key feature data; B5. Determine the supplier multimodal feature dataset based on the structured numerical feature data subset, the semi-structured semantic feature data subset, and the unstructured key feature data subset; B6. Perform weighted fusion processing on the set of credibility scores and the supplier multimodal feature dataset to obtain the supplier dual-weighted fusion feature dataset.

[0049] The set of credibility scores may include one or more credibility scores, which can refer to the score obtained by quantifying the credibility of each piece of calibrated multi-source heterogeneous data. It should be noted that the credibility score can reflect the authority, timeliness, and relevance to credit assessment of the data source, and this application does not impose any restrictions on this.

[0050] Specifically, based on the credibility assessment indicators (such as the qualifications of the source institution, the data generation time, and the degree of matching with the assessment dimensions) in the pre-set supplier credit assessment knowledge base, a weighted scoring method can be used to quantify and score each piece of calibrated multi-source heterogeneous data. For example, the credibility of financial institution data is 80 points, and anonymous evaluation is 30 points. All scores can be aggregated to form the aforementioned credibility score set.

[0051] The structured numerical feature data subset may include one or more structured numerical feature data, which may refer to numerical feature data extracted from structured data (such as financial statements, order forms, etc.) in calibration multi-source heterogeneous data, such as debt-to-equity ratio, on-time delivery rate, etc.

[0052] Specifically, field parsing technology can be used to extract numerical indicators corresponding to preset evaluation dimensions from structured data (such as Excel spreadsheets and database tables), such as accounts receivable turnover rate and the number of times contracts have been fulfilled in the past six months. Furthermore, these indicators can be categorized and organized according to their types to form and obtain the aforementioned subset of structured numerical feature data.

[0053] The subset of semi-structured semantic feature data may include one or more semi-structured semantic feature data. These semi-structured semantic feature data may refer to semantic features extracted from semi-structured data (such as contract texts, email records, etc.) in calibrated multi-source heterogeneous data, such as descriptions of contract breach clauses and records of cooperation disputes.

[0054] Specifically, natural language processing techniques (such as word segmentation, entity recognition, and sentiment analysis) can be used to extract key semantic information from semi-structured text, such as the proportion of penalties for delayed delivery and evaluation of cooperation satisfaction. The semantic information can be transformed into quantifiable feature vectors to form the aforementioned subset of semi-structured semantic feature data.

[0055] The subset of unstructured key feature data may include one or more unstructured key feature data. These unstructured key feature data may refer to key features extracted from unstructured data in calibration multi-source heterogeneous data (such as qualification certificate images, production site video frames, etc.), such as qualification validity period, equipment operating status, etc.

[0056] Specifically, optical character recognition technology can be used to extract text information from image data, such as the validity period and certification body on a qualification certificate. For video data, key status features, such as equipment operating parameters, can be extracted through frame analysis. The extracted information can then be standardized to form the aforementioned subset of unstructured key feature data.

[0057] The supplier multimodal feature dataset may include one or more supplier multimodal feature data sets. This dataset can refer to comprehensive feature data formed by integrating structured numerical feature data, semi-structured semantic feature data, and unstructured key feature data. It is understood that this supplier multimodal feature data can cover credit-related features of different data types.

[0058] Specifically, features from the three feature data subsets (i.e., structured numerical feature data subset, semi-structured semantic feature data subset, and unstructured key feature data subset) can be categorized according to evaluation dimensions (such as performance capability and compliance), and duplicate features can be further removed while retaining unique features under each modality, thereby forming a supplier multimodal feature dataset containing multiple types of features.

[0059] The supplier dual-weighted fusion feature dataset may include one or more supplier dual-weighted fusion feature data sets. These feature data sets refer to the features obtained by fusing the supplier multimodal feature dataset with both credibility scores and feature dimension importance weights. It should be noted that each supplier dual-weighted fusion feature data set can simultaneously carry data credibility weights and dimension category weights.

[0060] Specifically, we can first combine the credibility scores of the corresponding data in the credibility score set to weight and merge similar features in the supplier multimodal feature dataset. For example, for the same "delivery timeliness" feature from three data sources, with corresponding credibility scores of 80, 75, and 60 respectively, we can calculate a unified delivery timeliness feature value by weighting according to the credibility score proportion. Then, we can assign dimension category weights to each feature after merging. For example, we can predetermine the importance of each evaluation dimension, such as 35% for financial dimension features, 40% for fulfillment capability dimension features, and 25% for compliance dimension features. The dual-weight feature value of each feature is obtained by "merged feature value × corresponding dimension category weight". For features that are not of the same type, we can directly retain the dual-weight label and finally integrate them to form the above-mentioned supplier dual-weight fusion feature dataset.

[0061] In this embodiment, data quality differences are distinguished by credible quantification, features are extracted by modality to overcome the limitations of heterogeneous data formats, and multiple types of features are integrated to form a comprehensive supplier multimodal feature dataset. Then, dual-weighted fusion is used to balance data credibility and feature importance. The final output supplier dual-weighted fusion feature dataset retains the richness of multi-source data and strengthens the influence of high-quality features through the weight mechanism. This provides an accurate and comprehensive feature foundation for subsequent dynamic status updates and credit assessment, effectively improving the reliability and pertinence of the assessment process.

[0062] In one possible implementation, during the weighted fusion process, features in the supplier multimodal feature dataset can first be semantically aligned to obtain a supplier aligned feature dataset. Then, the aligned features are classified according to the credit assessment dimension to form a supplier credit classification feature dataset. The classified features of the same category are then weighted and fused together with a set of credibility scores to obtain a supplier credit category fused feature dataset. Finally, cross-modal non-category feature weights are assigned to the category fused features to ultimately obtain a supplier dual-weight fused feature dataset. Specifically, a method for obtaining a supplier dual-weight fused feature dataset by performing weighted fusion processing based on the set of credibility scores and the supplier multimodal feature dataset may include: C1. Perform semantic alignment processing on each supplier multimodal feature data in the supplier multimodal feature dataset to obtain a supplier aligned feature dataset; C2. Perform classification processing on each supplier alignment feature data in the supplier alignment feature dataset based on the credit assessment dimension to obtain the supplier credit classification feature dataset; C3. Perform weighted fusion processing of similar features based on the set of credibility scores and the set of supplier credit classification features to obtain a supplier credit similar fusion feature dataset; C4. Based on the supplier credit similar fusion feature dataset, perform cross-modal non-similar feature weight allocation processing to obtain the supplier dual-weight fusion feature dataset.

[0063] The supplier alignment feature dataset may include one or more supplier alignment feature data sets. These supplier alignment feature data sets refer to features that, after semantic alignment processing, are consistent in expression and semantically unambiguous. It should be noted that performing semantic alignment processing on each supplier multimodal feature data set can eliminate semantic differences between different modal features.

[0064] Specifically, based on the standard semantic rules in the pre-set supplier credit assessment knowledge base, synonymous features in multimodal feature data, such as payment cycle and payment period, equipment integrity rate and equipment availability rate, can be uniformly represented and transformed to ensure that features with the same meaning use consistent names and definitions, thereby forming the aforementioned supplier alignment feature dataset.

[0065] Credit assessment dimensions can refer to pre-defined core directions for supplier credit assessment, such as financial health, performance capability, compliant operation, and production stability. This application does not impose any restrictions on these dimensions.

[0066] The supplier credit classification feature dataset may include one or more supplier credit classification feature data, which can refer to feature data categorized according to credit assessment dimensions. It should be noted that each credit assessment dimension may contain all aligned features related to that dimension.

[0067] Specifically, based on the preset credit assessment dimensions, such as the financial health dimension which can include features like debt-to-equity ratio and cash flow, and the performance capability dimension which can include features like on-time delivery rate and order completion rate, the features in the supplier alignment feature dataset can be classified by dimension, thus forming a supplier credit classification feature dataset divided by dimension.

[0068] Similar features refer to features with the same meaning within the same credit assessment dimension in a supplier credit classification feature dataset. For example, the "debt-to-equity ratio" feature from different data sources within the financial health dimension. A supplier credit similar fusion feature dataset can include one or more similar fusion feature datasets. These similar fusion feature datasets refer to the feature data formed by weighted merging of similar features with credibility scores. It should be noted that after weighted fusion of each similar feature, only one unified feature value calculated based on the credibility weight is retained under the corresponding dimension. This feature value already integrates the credibility differences from the original multi-source data, and there is no need to separately retain multiple original credibility scores.

[0069] Specifically, in each dimension of the supplier credit classification feature dataset, features of the same type with the same meaning are selected, and the credibility scores of the corresponding features in the credibility score set are called. For example, feature A has a feature value of 85 and a credibility score of 90 from data source 1, and a feature value of 78 and a credibility score of 70 from data source 2. First, the weight ratio of the credibility scores of the same type of features is calculated (weight of data source 1 = 90 ÷ (90 + 70) = 56.25%, weight of data source 2 = 70 ÷ (90 + 70) = 43.75%). Then, the feature values ​​are weighted according to the weight ratio (feature fusion value = 85 × 56.25% + 78 × 43.75%). The calculation result is used as the unified fusion value of the same type of feature. After merging the fusion values ​​of all the same type of features, the original credibility scores and feature values ​​of the different types of features are retained, which can form the above-mentioned supplier credit same type fusion feature dataset.

[0070] Cross-modal dissimilar features can refer to features under different credit assessment dimensions within a supplier credit-related fusion feature dataset, such as "debt-to-equity ratio" in the financial health dimension and "on-time delivery rate" in the performance capability dimension. A supplier dual-weighted fusion feature dataset can refer to the feature set formed after assigning dimensional importance weights to cross-modal dissimilar features. Each supplier dual-weighted fusion feature in this set can carry a credibility-weighted fusion value and dimensional weights.

[0071] It should be noted that the aforementioned similar features, after weighted fusion, can be categorized into the corresponding credit assessment dimension as a single feature, and together with features from other dimensions (regardless of whether they have undergone similar fusion), constitute cross-modal dissimilar features. In other words, cross-modal dissimilar features can include similar fused features after weighted processing of similar features, i.e., a single feature resulting from the weighted merging of multiple original multimodal features with the same meaning under the same credit assessment dimension; and features in the supplier credit classification feature dataset that originally belonged to different credit assessment dimensions and had different meanings. These features were not processed by similar fusion and were directly retained as original single features.

[0072] Specifically, based on the preset importance ratio of each credit assessment dimension (such as 35% for financial health, 40% for performance capability, and 25% for compliance), corresponding dimension weights can be assigned to features under each dimension in the supplier credit fusion feature dataset, and all features can be further integrated to form the aforementioned supplier dual-weight fusion feature dataset.

[0073] In this embodiment, semantic alignment eliminates semantic ambiguity of multimodal features, classifies features according to credit assessment dimensions to clarify their attribution, and combines credibility to weighted fusion of similar features to ensure data quality. Furthermore, the dimensional weight allocation of cross-modal dissimilar features reflects the assessment focus. The resulting supplier dual-weighted fusion feature dataset not only ensures data reliability through credibility weights but also highlights the core assessment through dimensional weights. This effectively integrates the value of multi-source heterogeneous data, providing a structured and high-quality feature foundation for subsequent accurate supplier credit assessment and improving the scientific rigor and relevance of the assessment results.

[0074] In one possible implementation, during the dynamic state update process, the supplier dual-weight fusion feature data can first be split into time-series slices to obtain a static basic feature slice subset and a historical dynamic incremental feature slice subset; then, real-time multi-source heterogeneous datasets of the suppliers to be evaluated are collected and a real-time dynamic incremental feature slice subset is extracted; the historical dynamic incremental feature slices are subjected to timeliness detection to obtain a time-sensitive dynamic incremental feature slice subset; the time-sensitive dynamic incremental feature slice subset and the real-time dynamic incremental feature slice subset are fused using freshness weighting to obtain an updated dynamic incremental feature slice subset; finally, the static basic feature slice subset and the updated dynamic incremental feature slice subset are fused to form the supplier dynamic feature dataset. Specifically, a method for obtaining a supplier dynamic feature dataset by performing dynamic state update processing based on the supplier dual-weight fusion feature dataset may include: D1. Perform time-series slicing on each supplier dual-weighted fusion feature data in the supplier dual-weighted fusion feature dataset to obtain a static basic feature slice subset and a historical dynamic incremental feature slice subset. D2. Real-time collection of multi-source heterogeneous datasets related to the supplier to be evaluated; D3. Based on the real-time multi-source heterogeneous dataset, determine the real-time dynamic incremental feature slice subset; D4. Perform timeliness detection processing on each historical dynamic incremental feature slice in the historical dynamic incremental feature slice subset to obtain a timeliness dynamic incremental feature slice subset. D5. Perform freshness-weighted fusion processing on the time-sensitive dynamic incremental feature slice subset and the real-time dynamic incremental feature slice subset to obtain the updated dynamic incremental feature slice subset. D6. The static basic feature slice subset and the updated dynamic incremental feature slice subset are fused to obtain the supplier dynamic feature dataset.

[0075] The static basic feature slice subset may include one or more static basic feature slices. These static basic feature slices can refer to feature slices that are stable over a long period and do not change significantly over time, extracted from the dual-weighted fusion features, such as supplier registered capital and core qualification levels. The historical dynamic incremental feature slice subset may include one or more historical dynamic incremental feature slices. These historical dynamic incremental feature slices can refer to historical feature slices that change dynamically over time, extracted from the dual-weighted fusion features, such as on-time delivery rate and quarterly financial indicators over the past three months.

[0076] Specifically, the dual-weighted fused feature data can be sliced ​​based on the time dimension to select features that have not changed significantly in recent years as static basic feature slices, and dynamically changing features can be split according to time period (such as week, month, quarter) as historical dynamic incremental feature slices to form the above two subsets respectively. This application does not limit this.

[0077] The real-time multi-source heterogeneous dataset may include one or more real-time multi-source heterogeneous datasets. These real-time multi-source heterogeneous datasets may refer to the latest data collected in real time related to the supplier to be evaluated, from multiple channels and in various formats, such as daily transaction records, real-time production status data, and the latest public opinion information. This application does not limit this.

[0078] Specifically, technologies such as interface integration, real-time web crawling, and sensor data acquisition can be used to obtain the latest data of suppliers to be evaluated in real time from channels such as supplier enterprise resource planning (ERP) systems, third-party trading platforms, and industry regulatory databases, and summarize them to form a real-time multi-source heterogeneous dataset.

[0079] The real-time dynamic incremental feature slice subset may include one or more real-time dynamic incremental feature slices. These real-time dynamic incremental feature slices may refer to dynamic feature slices extracted from real-time multi-source heterogeneous datasets that reflect the latest status of the supplier, such as the order fulfillment progress of the day, the latest accounts receivable data, etc.

[0080] It should be noted that the details regarding determining the real-time dynamic incremental feature slice subset based on the real-time multi-source heterogeneous dataset can be found in the aforementioned detailed description of determining the historical dynamic incremental feature slice subset based on the historical multi-source heterogeneous dataset, and will not be repeated here. Specifically, the real-time multi-source heterogeneous dataset can be subjected to feature extraction and standardization processing with reference to the dimensions and format of the historical dynamic incremental features to form a real-time dynamic incremental feature slice subset with dimensions consistent with the historical dynamic incremental feature slices.

[0081] The time-sensitive dynamic incremental feature slice subset may include one or more time-sensitive dynamic incremental feature slices. These time-sensitive dynamic incremental feature slices may refer to the dynamic incremental feature slices that still have reference value after time-sensitive detection, after removing expired feature slices or clearing invalid feature slices.

[0082] Specifically, based on preset timeliness rules, such as a one-quarter validity period for financial data and a one-month validity period for performance records, the generation time of historical dynamic incremental feature slices can be detected to retain feature slices that have not exceeded the timeliness threshold and remove feature slices that have exceeded the timeliness threshold, thereby forming the aforementioned subset of timeliness-sensitive dynamic incremental feature slices. Optionally, a sliding window mechanism can be used, such as setting a fixed time window (e.g., a one-month window, a three-month window) to filter timeliness-sensitive feature slices in real time, to further improve the accuracy and dynamic adaptability of timeliness detection.

[0083] It should be noted that freshness can be used to indicate the temporal freshness of a feature slice. Understandably, the higher the temporal freshness of a feature slice, the greater its reference value for the current credit assessment process; for example, the freshness of real-time data can be higher than data from 3 days ago. Conversely, the lower the temporal freshness of a feature slice, the less its reference value for the current credit assessment process. This application does not impose any restrictions on this.

[0084] The updated dynamic incremental feature slice subset may include one or more updated dynamic incremental feature slices. These updated dynamic incremental feature slices may refer to feature slices that reflect the latest dynamic trends after fusing time-sensitive dynamic incremental feature slices and real-time dynamic incremental feature slices.

[0085] Specifically, freshness weights can be assigned to time-sensitive dynamic incremental feature slices and real-time dynamic incremental feature slices. For example, the freshness weight of real-time data can be 80%, the freshness weight of data within one week can be 50%, and the freshness weight of data within one month can be 30%. These weighted features can be fused according to the same feature dimension. For example, for the freshness weighted fusion of "delivery timeliness", the updated dynamic incremental feature slice of "delivery timeliness" can be 80% of the real-time dynamic incremental feature slice and 50% of the time-sensitive dynamic incremental feature slice (such as the dynamic incremental feature slice within one week) to form the above-mentioned subset of updated dynamic incremental feature slices.

[0086] Furthermore, the comprehensive feature dataset formed by merging static basic feature slices and updated dynamic incremental feature slices, which contains both stable basic information and reflects the latest dynamic changes, is the aforementioned supplier dynamic feature dataset. Specifically, the subset of static basic feature slices and the subset of updated dynamic incremental feature slices can be integrated according to the evaluation dimension to retain the long-term stability of static features and the timeliness of dynamic features, thereby forming a supplier dynamic feature dataset that can cover the full-dimensional status of suppliers.

[0087] In this embodiment, the supplier's dual-weighted fusion feature data is split into static basic features and historical dynamic incremental features. Real-time dynamic incremental features are extracted by combining the latest data collected in real time. Valid historical features are screened through timeliness detection and then fused with the real-time features using a freshness-weighted fusion method. Finally, the static and updated dynamic features are integrated to form a supplier dynamic feature dataset. This approach not only preserves the stability of the supplier's core basic information but also accurately captures its latest operational status through a dynamic update mechanism. It effectively eliminates the interference of outdated and invalid data on the evaluation results, significantly improving the timeliness and accuracy of the feature data. This provides high-quality data support for subsequent dynamic and scientific supplier credit evaluation, ensuring that the credit evaluation results can reflect the supplier's current credit level in real time and better meet the actual needs of risk control and cooperation decision-making in enterprise supply chain management.

[0088] For examples consistent with the above embodiments, please refer to... Figure 3 , Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application, such as... Figure 3 As shown, it includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps. Obtain historical, multi-source, heterogeneous datasets related to the suppliers to be evaluated; Based on a pre-defined supplier credit assessment knowledge base, dynamic logical calibration is performed on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset. Based on the calibration multi-source heterogeneous dataset, a confidence-based multimodal fusion process is performed to obtain a supplier dual-weighted fusion feature dataset; The supplier dynamic feature dataset is obtained by performing dynamic state update processing based on the supplier dual-weight fusion feature dataset. The supplier credit assessment is performed on the supplier to be evaluated based on the supplier dynamic feature dataset to obtain the target supplier credit assessment result.

[0089] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0090] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0091] For those consistent with the above, please refer to Figure 4 , Figure 4 This application provides a schematic diagram of a supplier credit assessment device based on multi-source heterogeneous data. For example... Figure 4 As shown, the device includes: Acquisition unit 101 is used to acquire historical multi-source heterogeneous datasets related to the supplier to be evaluated; The first processing unit 102 is used to perform dynamic logical calibration processing on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset based on a preset supplier credit assessment exclusive knowledge base, so as to obtain a calibrated multi-source heterogeneous dataset. The second processing unit 103 is used to perform confidence-based multimodal fusion processing on the calibration multi-source heterogeneous dataset to obtain a supplier dual-weight fusion feature dataset. The third processing unit 104 is used to perform dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain the supplier dynamic feature dataset. The fourth processing unit 105 is used to perform supplier credit assessment processing on the supplier to be evaluated based on the supplier dynamic feature dataset, and obtain the target supplier credit assessment result.

[0092] In one possible implementation, the first processing unit 102 is configured to perform dynamic logical calibration processing on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset based on a preset supplier credit assessment-specific knowledge base, to obtain a calibrated multi-source heterogeneous dataset, specifically for: Key information is extracted from each historical multi-source heterogeneous data in the aforementioned historical multi-source heterogeneous dataset to obtain a key information dataset. Based on a pre-defined supplier credit assessment knowledge base, term mapping is performed on each key information data in the key information dataset to obtain a standard key information dataset. By invoking the logical rules in the preset supplier credit assessment knowledge base, correlation verification is performed on each standard key information data in the standard key information dataset to obtain a correlation conflict dataset. Each related conflict data point in the related conflict dataset is subjected to hierarchical dynamic correction processing to obtain a calibrated multi-source heterogeneous dataset.

[0093] In one possible implementation, the second processing unit 103 is configured to perform confidence-based multimodal fusion processing on the calibration multi-source heterogeneous dataset to obtain a supplier dual-weighted fusion feature dataset, specifically for: The credibility quantification process is performed on each calibration multi-source heterogeneous data in the calibration multi-source heterogeneous dataset to obtain a set of credibility scores. Based on the aforementioned calibrated multi-source heterogeneous dataset, structured numerical feature extraction processing is performed to obtain a subset of structured numerical feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, semi-structured semantic feature extraction processing is performed to obtain a subset of semi-structured semantic feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, unstructured key feature extraction processing is performed to obtain a subset of unstructured key feature data; Based on the structured numerical feature data subset, the semi-structured semantic feature data subset, and the unstructured key feature data subset, a supplier multimodal feature dataset is determined; The supplier dual-weighted fused feature dataset is obtained by performing a weighted fusion process on the set of credibility scores and the supplier multimodal feature dataset.

[0094] In one possible implementation, the second processing unit 103 is configured to perform a weighted fusion process based on the confidence score set and the supplier multimodal feature dataset to obtain a supplier dual-weighted fusion feature dataset, specifically for: Semantic alignment is performed on each supplier multimodal feature data in the supplier multimodal feature dataset to obtain a supplier aligned feature dataset; Each supplier alignment feature in the supplier alignment feature dataset is classified based on the credit assessment dimension to obtain a supplier credit classification feature dataset. Based on the set of credibility scores and the set of supplier credit classification features, a weighted fusion process of similar features is performed to obtain a supplier credit similar fusion feature dataset. Based on the supplier credit similar fusion feature dataset, cross-modal dissimilar feature weight allocation processing is performed to obtain the supplier dual-weight fusion feature dataset.

[0095] In one possible implementation, the third processing unit 104 is used to perform dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain a supplier dynamic feature dataset, specifically for: Each supplier dual-weight fusion feature data in the supplier dual-weight fusion feature dataset is split into time-series slices to obtain a static basic feature slice subset and a historical dynamic incremental feature slice subset. Real-time collection of multi-source heterogeneous datasets related to the supplier to be evaluated; Based on the real-time multi-source heterogeneous dataset, determine the real-time dynamic incremental feature slice subset; Perform timeliness detection processing on each historical dynamic incremental feature slice in the historical dynamic incremental feature slice subset to obtain a timeliness dynamic incremental feature slice subset. The updated dynamic incremental feature slice subset is obtained by performing freshness-weighted fusion processing on the time-sensitive dynamic incremental feature slice subset and the real-time dynamic incremental feature slice subset. The supplier dynamic feature dataset is obtained by fusing the static basic feature slice subset and the updated dynamic incremental feature slice subset.

[0096] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the supplier credit assessment methods based on multi-source heterogeneous data as described in the above method embodiments.

[0097] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the supplier credit assessment methods based on multi-source heterogeneous data as described in the above method embodiments.

[0098] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0099] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0100] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0101] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0102] Furthermore, the functional units in the various embodiments of the application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0103] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0104] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0105] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A supplier credit assessment method based on multi-source heterogeneous data, characterized in that, The supplier credit assessment method based on multi-source heterogeneous data includes: Obtain historical, multi-source, heterogeneous datasets related to the suppliers to be evaluated; Based on a pre-defined supplier credit assessment knowledge base, dynamic logical calibration is performed on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset to obtain a calibrated multi-source heterogeneous dataset. Based on the calibration multi-source heterogeneous dataset, a confidence-based multimodal fusion process is performed to obtain a supplier dual-weighted fusion feature dataset; The supplier dynamic feature dataset is obtained by performing dynamic state update processing based on the supplier dual-weight fusion feature dataset. The supplier credit assessment is performed on the supplier to be evaluated based on the supplier dynamic feature dataset to obtain the target supplier credit assessment result.

2. The supplier credit assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The process involves dynamically calibrating each historical multi-source heterogeneous data point in the historical multi-source heterogeneous dataset based on a pre-defined supplier credit assessment knowledge base, resulting in a calibrated multi-source heterogeneous dataset, including: Key information is extracted from each historical multi-source heterogeneous data in the aforementioned historical multi-source heterogeneous dataset to obtain a key information dataset. Based on a pre-defined supplier credit assessment knowledge base, term mapping is performed on each key information data in the key information dataset to obtain a standard key information dataset. By invoking the logical rules in the preset supplier credit assessment knowledge base, correlation verification is performed on each standard key information data in the standard key information dataset to obtain a correlation conflict dataset. Each related conflict data point in the related conflict dataset is subjected to hierarchical dynamic correction processing to obtain a calibrated multi-source heterogeneous dataset.

3. The supplier credit assessment method based on multi-source heterogeneous data according to claim 2, characterized in that, The step of performing credibility-based multimodal fusion processing on the calibrated multi-source heterogeneous dataset to obtain a supplier dual-weighted fusion feature dataset includes: The credibility quantification process is performed on each calibration multi-source heterogeneous data in the calibration multi-source heterogeneous dataset to obtain a set of credibility scores. Based on the aforementioned calibrated multi-source heterogeneous dataset, structured numerical feature extraction processing is performed to obtain a subset of structured numerical feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, semi-structured semantic feature extraction processing is performed to obtain a subset of semi-structured semantic feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, unstructured key feature extraction processing is performed to obtain a subset of unstructured key feature data; Based on the structured numerical feature data subset, the semi-structured semantic feature data subset, and the unstructured key feature data subset, a supplier multimodal feature dataset is determined; The supplier dual-weighted fused feature dataset is obtained by performing a weighted fusion process on the set of credibility scores and the supplier multimodal feature dataset.

4. The supplier credit assessment method based on multi-source heterogeneous data according to claim 2, characterized in that, The step of performing a weighted fusion process based on the confidence score set and the supplier multimodal feature dataset to obtain the supplier dual-weighted fusion feature dataset includes: Semantic alignment is performed on each supplier multimodal feature data in the supplier multimodal feature dataset to obtain a supplier aligned feature dataset; Each supplier alignment feature in the supplier alignment feature dataset is classified based on the credit assessment dimension to obtain a supplier credit classification feature dataset. Based on the set of credibility scores and the set of supplier credit classification features, a weighted fusion process of similar features is performed to obtain a supplier credit similar fusion feature dataset. Based on the supplier credit similar fusion feature dataset, cross-modal dissimilar feature weight allocation processing is performed to obtain the supplier dual-weight fusion feature dataset.

5. The supplier credit assessment method based on multi-source heterogeneous data according to any one of claims 1-4, characterized in that, The dynamic state update process based on the supplier dual-weight fusion feature dataset yields the supplier dynamic feature dataset, including: Each supplier dual-weight fusion feature data in the supplier dual-weight fusion feature dataset is split into time-series slices to obtain a static basic feature slice subset and a historical dynamic incremental feature slice subset. Real-time collection of multi-source heterogeneous datasets related to the supplier to be evaluated; Based on the real-time multi-source heterogeneous dataset, determine the real-time dynamic incremental feature slice subset; Perform timeliness detection processing on each historical dynamic incremental feature slice in the historical dynamic incremental feature slice subset to obtain a timeliness dynamic incremental feature slice subset. The updated dynamic incremental feature slice subset is obtained by performing freshness-weighted fusion processing on the time-sensitive dynamic incremental feature slice subset and the real-time dynamic incremental feature slice subset. The supplier dynamic feature dataset is obtained by fusing the static basic feature slice subset and the updated dynamic incremental feature slice subset.

6. A supplier credit assessment device based on multi-source heterogeneous data, characterized in that, The device includes: The acquisition unit is used to acquire historical multi-source heterogeneous datasets related to the suppliers to be evaluated. The first processing unit is used to perform dynamic logical calibration processing on each historical multi-source heterogeneous data in the historical multi-source heterogeneous dataset based on a preset supplier credit assessment exclusive knowledge base, so as to obtain a calibrated multi-source heterogeneous dataset. The second processing unit is used to perform confidence-based multimodal fusion processing on the calibration multi-source heterogeneous dataset to obtain a supplier dual-weight fusion feature dataset. The third processing unit is used to perform dynamic state update processing based on the supplier dual-weight fusion feature dataset to obtain the supplier dynamic feature dataset. The fourth processing unit is used to perform supplier credit assessment processing on the supplier to be evaluated based on the supplier dynamic feature dataset, and obtain the target supplier credit assessment result.

7. The supplier credit assessment device based on multi-source heterogeneous data according to claim 6, characterized in that, The process of dynamically calibrating each historical multi-source heterogeneous data point in the historical multi-source heterogeneous dataset based on a pre-defined supplier credit assessment knowledge base to obtain a calibrated multi-source heterogeneous dataset is specifically used for: Key information is extracted from each historical multi-source heterogeneous data in the aforementioned historical multi-source heterogeneous dataset to obtain a key information dataset. Based on a pre-defined supplier credit assessment knowledge base, term mapping is performed on each key information data in the key information dataset to obtain a standard key information dataset. By invoking the logical rules in the preset supplier credit assessment knowledge base, correlation verification is performed on each standard key information data in the standard key information dataset to obtain a correlation conflict dataset. Each related conflict data point in the related conflict dataset is subjected to hierarchical dynamic correction processing to obtain a calibrated multi-source heterogeneous dataset.

8. The supplier credit assessment device based on multi-source heterogeneous data according to claim 7, characterized in that, The step of performing a confidence-based multimodal fusion process on the calibrated multi-source heterogeneous dataset to obtain a supplier dual-weighted fusion feature dataset is specifically used for: The credibility quantification process is performed on each calibration multi-source heterogeneous data in the calibration multi-source heterogeneous dataset to obtain a set of credibility scores. Based on the aforementioned calibrated multi-source heterogeneous dataset, structured numerical feature extraction processing is performed to obtain a subset of structured numerical feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, semi-structured semantic feature extraction processing is performed to obtain a subset of semi-structured semantic feature data; Based on the aforementioned calibrated multi-source heterogeneous dataset, unstructured key feature extraction processing is performed to obtain a subset of unstructured key feature data; Based on the structured numerical feature data subset, the semi-structured semantic feature data subset, and the unstructured key feature data subset, a supplier multimodal feature dataset is determined; The supplier dual-weighted fused feature dataset is obtained by performing a weighted fusion process on the set of credibility scores and the supplier multimodal feature dataset.

9. A terminal, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the supplier credit assessment method based on multi-source heterogeneous data as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the supplier credit assessment method based on multi-source heterogeneous data as described in any one of claims 1-5.

Images